What I mean by 'machine audits'

This is a concept that doesn't quite exist yet, so I've employed this generic term instead of blundering about with terms like "parse" and "validate."

What I mean is that when you've created a cooperative system of coding and declaring and storing semantic meaning in directories and meaning models, it will be possible to create software that crawls semantic sites and runs various routines against its semantic declarations.

• Is the XML well-formed and valid? Is the schema correct and consistent for the declaration?
• Do the claims in the triples agree or contradict?
• Can sample XML be parsed?
• Are the terms contained in the meaning model consistent with the site's declarations about controlled vocabularies, etc.?
• Is the content producer complying with "best practices" standards put forward by any certifying bodies?

Again, I see no reason to police whether people are lying or misleading or fair or unfair. If your content can pass these kinds of audits, then whether or not you're producing shoddy content will be glaringly apparent.

Having to do this entirely by hand, with human auditors, is a non-answer.